Success Semantics, Reinforcing Satisfaction, and Sensory Inclinations

Abstract

Success semantics holds, roughly, that what it is for a state of an agent to be a belief that P is for it to be disposed to combine with her desires to cause behaviour that would fulfill those desires if P. J. T. Whyte supplements this with an account of the contents of an agent's “basic desires” to provide an attractive naturalistic theory of mental content. We argue that Whyte's strategy can avoid the objections raised against it by restricting “basic desires” to sensory inclinations that cause us to do things independently of our beliefs about their contents.

Résumé

La sémantique du succès soutient, en gros, que ce qu'il faut pour qu'un état d'un agent soit une croyance en P, c'est que cet état soit disposé à se combiner avec ses désirs pour provoquer un comportement qui répondrait à ces désirs si P. J. T. Whyte complète cela par un compte-rendu du contenu des « désirs de base » d'un agent pour fournir une théorie naturaliste attrayante du contenu mental. Nous soutenons que la stratégie de Whyte peut éviter les objections qui lui sont faites en restreignant les « désirs de base » aux inclinations sensorielles qui nous font agir indépendamment de nos croyances sur leur contenu.

1. Whyte's Success Semantics

Most generally and roughly, success semantics holds that

(R) What it is for a state of an agent B to be a belief that P is for B to be disposed to combine with the agent's desires to cause behaviour that would fulfill those desires if P.

For example, what makes it the case that some (e.g., neural) state of Mary is a belief that the food in front of her is poisoned is that this state is disposed to combine with her desires to cause behaviour that would fulfill those desires if it was poisoned. The state would, for instance, combine with Mary's desire to eat such food only if it is not poisoned to cause her to avoid eating it.

Functionalist accounts seek to explain the nature of mental states in terms of their causal dispositions. Success semantics is a functionalist account not only of what it is for something to be an instance of a general kind of mental state — like belief — but what it is for it to have a particular intentional content — or be a belief that P.¹ Success semantics thus has the general virtues of functionalism of explaining how mental states can be multiply realized by different sorts of physical states without being something “over and above” the physical.² For instance, Mike's belief that sharks are present might be realized by the dispositions of certain of his cortical neurons, while beliefs with the same content in Joe, a seabird, an octopus, an alien observer, and an AI system of the future might be realized respectively by slightly different neurons in Joe's cortex, the bird's homologous but more nuclear cerebrum, the octopus's ganglia, the alien's fluid sacks, and the AI's silicon hardware.³ R plausibly captures the key causal dispositions common to all these physical states in virtue of which we should see them as realizing beliefs with the same or at least relevantly similar content.⁴

While Frank Ramsey (1990) is typically credited with this basic idea of success semantics, J. T. Whyte (1990, 1991) has developed it into a comprehensive naturalistic theory of mental states’ intentionality or content in non-intentional terms. Since R explains beliefs’ contents in terms of the contents of desires, such a theory needs a naturalistic account of desires’ contents. As Whyte observes, a plausible parallel functionalist account of desires’ contents in terms of their causal dispositions is

(F) What it is for a state of an agent D to be a desire that O (i.e., that will be fulfilled by outcome O) is for D to be disposed to combine with the agent's beliefs to cause behaviour that would bring it about that O if those beliefs were true (i.e., if their contents were to obtain).

For instance, the causal essence of a desire to eat food of a given kind only if it is not poisoned seems to be a disposition to combine with one's beliefs (e.g., that the food of that kind in front of one is poisoned) to cause one to do things (like avoid eating the food) that would bring it about that one eats food of that kind only if it is not poisoned if those beliefs were true (and e.g., the food in front of one is poisoned).

Unfortunately, as Whyte notes, a theory of content comprised of nothing but R and F seems viciously circular. The problem is not simply that R and F describe beliefs and desires as interlocking dispositional states, but rather that too little has been said about these interlocking dispositions in non-intentional terms for them to constitute a substantive naturalistic theory of content. One can use the Ramsey-Carnap-Lewis method to give a non-circular account of entities described by interlocking dispositions by replacing the names of these entities with bound variables and saying that there exists a set of entities that relate to each other in the relevant ways (Braddon-Mitchell & Jackson, 2007; Lewis, 1970).⁵ But if one tries to use this approach to explain the contents of beliefs and desires using only R and F, all one can say is that there are two kinds of states which are disposed to combine to produce behaviour that fulfills or brings about the content of one if the other is true or such that its content obtains. But this does not seem to say enough to explain the properties of a state's being fulfilled or such that its content is brought about, or its being true or such that its content obtains.

Still, as Whyte suggests, if we can give a naturalistic account of the contents of some desires that does not refer to the contents of beliefs, we can then use R and F to build upon this foundation to give an account of the contents of beliefs and other desires. Whyte's proposal is in essence that there is a set of “basic desires,” the contents of which we can account for naturalistically in terms of what reinforcingly satisfies them:

(S) What it is for a state of an agent D to be a basic desire that O is for it to be the case that O would (i) cause D to “go away” or cease exerting causal influence on the agent's behaviour, (ii) in a way that would reinforce the agent's disposition to act in the way that led to O when D is next present or active.⁶

For instance, Whyte thinks that desires like those to possess and eat cherries are basic desires: possessing cherries causes one's desire to possess them to cease influencing one's conduct, and eating cherries causes one's desire to eat them to cease influencing one's conduct. Of course, desires can cease influencing one's conduct through the obtaining of things other than their content — such as one's receiving a hard blow to the stomach. But when one lacks a desire for a blow to the stomach, one's receiving a blow does not tend to cause one to repeat whatever led to one's receiving it. By contrast, it is plausibly essential to basic desires or inclinations like those to possess or eat cherries that one has a tendency to become more likely to do whatever it was that led to one's possessing or eating them if one is motivated by those desires in the future.⁷

2. Problems for the Reinforcing Satisfaction Account

Whyte thinks, however, that S as it stands fails to provide either necessary or sufficient conditions for an outcome to be the content of, or what fulfills, a basic desire. He claims that an outcome O can be the content of a basic desire D, even though it is not the case that O would (i) cause D to go away (ii) in a way that reinforces the behaviour that led to O. For instance, Whyte suggests that Prema's desire to possess cherries could be fulfilled by someone's putting cherries into her handbag when she is not looking. Especially if Prema never comes to know when or how she got the cherries (or even that she has them), it seems that this fulfillment of her desire to possess cherries could fail to reinforce any of the behaviours (e.g., her approaching the individual who surreptitiously gave them to her) that led to its fulfillment.⁸

Whyte also claims that it is possible for outcome O to cause basic desire D to (i) go away (ii) in a way that reinforces the behaviour that led to O, without O actually being the content of D. For instance, suppose that Prema cannot distinguish genuine cherries from similar-tasting imitation cherries, but that she nonetheless desires to eat only genuine cherries, say because she believes that they are healthier. Whyte suggests that Prema's eating imitation cherries could (i) cause her desire to eat genuine cherries to go away, (ii) in a way that reinforces her tendency to do whatever led to her eating imitation cherries — such as her ordering “cherries” from someone she thinks is selling her genuine ones — if she is moved by her desire for genuine cherries in the future.

In response to these problems, Whyte (1991, pp. 70–73) proposes a modification of S, the basic idea of which is that a basic desire's content is what would reinforcingly satisfy it under “normal conditions.” Thus, a state of Prema's can be a basic desire to possess cherries despite its not being reinforcingly satisfied by possessing cherries she does not notice, and a state of Prema's can be a basic desire to eat genuine cherries despite its being reinforcingly satisfied by imitation cherries, because her being unable to distinguish cherries’ absence from presence and genuineness from imitation is relevantly “abnormal.” Whyte notes, however, that it would be viciously circular to understand “normal” conditions as whatever conditions are such that under them a basic desire is reinforcingly satisfied just in case it is fulfilled. He also notes that it would be a mistake to identify normal conditions with whatever conditions are statistically typical, since it is possible for error to be statistically typical. Whyte proposes to understand normal conditions as those that would remain reinforcingly satisfying no matter how much the agent's perceptual capacities were improved, or

(S’): What it is for a state of an agent D to be a basic desire that O is for it to be the case that O would (i) cause D to go away (ii) in a way that would reinforce the agent's disposition to act in the way that led to O when D is next active, and (iii) (i) and (ii) would remain true no matter how much the agent's perceptual capacities were improved.

Anandi Hattiangadi has objected that S’ is viciously circular. As she puts it, “In order to decide what counts as an ‘improvement’ of my perceptual abilities, assumptions have to be made about what I want, which is ultimately circular” (Hattiangadi, 2007, p. 125). We do not think, however, that the circularity is as obvious as Hattiangadi appears to suppose. It might seem that Whyte could characterize perceptual improvements as something like increased abilities to discriminate among or respond differentially to different states of the world. Such an account of improvement would not obviously presuppose the content of the agent's desires.

That said, we think Hattiangadi is correct that the plausibility of S’ trades upon a kind of vicious circularity. It does not seem that the contents of an agent's desires should depend upon what would reinforcingly satisfy them if she underwent any arbitrary increase in discriminative ability. For instance, what would be true in fanciful scenarios such as Prema's gaining the ability to perceptually discriminate among subatomic particles does not seem relevant to the content of her desires like those to possess and eat cherries. For S’ to be a plausible explanation of the content of Prema's desires to possess and eat cherries, it needs instead to consider such things as what would be true if she were able to discriminate between present and absent cherries and between genuine and imitation cherries. But to distinguish these discriminations as relevant from others that are not seems to presuppose the contents of Prema's desires in a viciously circular way.

Moreover, an arbitrary increase in an agent's discriminative abilities might alter the contents of her desires. This would cause (i) and (ii) to cease to hold for some outcome O not because O was not what the agent desired, but because the changes mentioned in (iii) change what she desires. For example, Prema might currently have a desire to eat cherries that would be fulfilled by cherries with a wide range of acidity levels, but improving her range of gustatory discrimination would cause her to desire to eat only cherries with a very particular acidity level. It does not seem that Whyte's approach to solving the problems he raises with S allows him to distinguish between perceptual improvements that do as opposed to do not alter basic desires’ contents without viciously circular assumptions about what their contents actually are.

3. Basic Desires as Sensory Inclinations

We think, however, that there is an alternative version of S that avoids the problems that Whyte raises for it, which avoids appealing to “normal conditions” and thus also avoids the problems with S’. Our solution looks more carefully at the role that should be played by Whyte's “basic desires.” These should be simple motivational states of an agent, the causal tendencies of which do not depend upon her beliefs about their contents. But Whyte's examples of allegedly “basic” desires that make trouble for S are actually desires like those to possess cherries and to eat genuine as opposed to imitation cherries.⁹ Whether these desires are reinforcingly satisfied obviously depends upon whether the agent believes that their contents obtain.

But agents’ most phylogenetically and ontogenetically conserved motivational states do not seem to be like this. These motives include the urges, likes, and dislikes involved in (or at least accompanying) such sensory states as itches, hunger, thirst, pleasurable experiences, and affectively painful experiences.¹⁰ The objects of these sensory inclinations are internal states of the agent herself, such as her experiencing scratching, experiencing the ingestion of food or water, or her continuing or ceasing to have certain experiences. These motives are reinforcingly satisfied by the obtaining of the internal states of the agent that are their content, rather than the agent's beliefs or perceptions that these states obtain.¹¹

To the extent that sensory inclinations are motives for the agent to have certain subjective experiences, it is arguable that her introspective representations of whether they are fulfilled cannot be false. But it is also arguable that these experiences can occur in the absence of beliefs or representations about them, and when they do, they reinforcingly satisfy sensory inclinations for them. Many agents, such as human infants and some sentient non-human animals, arguably have sensory inclinations that are reinforcingly satisfied by the experiences that are their contents, although they lack the capacity to form representations to themselves about whether they are having these experiences. When we potentially reflective agents are too busy experiencing the world and acting upon it to indulge in navel-gazing reflection about how we feel, we are arguably identical in this respect to our younger and specifically different fellows.¹²

For instance, an infant, vole, or adult human who lacks time to reflect on what she is experiencing might like and have a sensory inclination to experience cherry-like taste. Having an experience of cherry-like taste will cause this inclination to (i) go away (ii) in a way that will tend to make her more likely in the future to do whatever caused her to have this experience the next time she is moved by the inclination. This can all be true without her ever having a tendency to represent to herself that she is experiencing cherry-like taste. Or consider a human adult who has a sensory inclination to experience cherry-like taste that she mistakes for a motivation to eat genuine as opposed to imitation cherries. This adult might discover that she is mistaken about her own motives by seeing how her cravings are reinforcingly satisfied by imitation just as much as by genuine cherries.

It seems then that we should offer S as an account of the content of sensory inclinations, which are reinforcingly satisfied by the states of the agent that are their content, rather than the content of Whyte's “basic desires” which are reinforcingly satisfied by the agent's beliefs that their contents obtain. We thus arrive at

(S*) What it is for a state of an agent SI to be a sensory inclination that internal state E of the agent obtains is for it to be the case that E would (i) cause SI to go away (ii) in a way that would reinforce the agent's disposition to act in the way that led to E when SI is next active.

4. Conclusion: A Foundation for Mental Content

With S* in place, we can use R to give an account of what it is for an initial set of “level 1” representational states to have the content that P in terms of their tendency to combine with the agent's sensory inclinations to cause behaviour that would fulfill her inclinations if P. As the most basic representations with which sensory inclinations combine to influence behaviour, these might actually be more plausibly understood as perceptions or sensory representations than beliefs (see Hall, 2008, pp. 532–533). Having done this, we can then use F to give an account of what it is for “level 2 motivations” to have the content that O in terms of their tendency to combine with level 1 representations to cause behaviour that would bring about O if the content of the level 1 representations were to obtain. From here we can iterate further, giving an account of what it is for “level 2 representations” to have the content that P in terms of their tendency to combine with level 1 motivations and level 2 motivations to cause behaviour that would fulfill them if P, and so on.¹³

What, however, if the states of an agent that are the objects of her sensory inclinations are experiences that themselves have representational contents, so that, for instance, a painful sensation she is inclined to terminate represents tissue damage at a particular bodily location (cf. Tye, 1995)? One possibility is that the contents of sensory or perceptual representations should be explained by a different theory, such as a causal covariation account (Tye, 1995, pp. 100–105), and using this together with S*'s account of the contents of sensory inclinations as a foundation, success semantics should seek to use R and F to explain only the contents of beliefs and desires proper.

Alternatively, we do not think that it would be viciously circular to use S*, R, and F to offer a fully general theory of mental content, including the representations of the objects of sensory inclinations. On this theory, the dispositions described by R and S* are interlocking, but we think that S* can still provide enough independent traction on the idea of what it is for a sensory inclination to be fulfilled or unfulfilled to make substantive sense of the behavioural dispositions that R appeals to in explaining the contents of representations that combine with these motives. This in turn allows for a substantive understanding of the dispositions that F appeals to in explaining the contents of motives that combine with these representations, and so on. For instance, S* explains what it is for an inclination of an agent to be fulfilled by the termination of her experience of tissue damage in terms of this termination reinforcingly satisfying the inclination. This allows us to explain what it is for the experience of tissue damage itself to involve a level 1 representation of tissue damage in terms of its combining with the inclination and other representations that would cause behaviour that would fulfill the inclination if these representations were accurate. For instance, the experience's representation of tissue damage would combine with the inclination to terminate the experience and the representation that grasping the limb where the tissue damage seems to be occurring will terminate the experience to cause behaviour like reaching to grasp the limb that would fulfill (i.e., reinforcingly satisfy) the inclination if the tissue damage is in the limb and grasping it will terminate the experience (and this is not, for example, a case of referred or phantom limb pain in which reaching for the limb will not terminate the pain). The fact that the experience which reinforcingly satisfies the inclination itself has a representational content of tissue damage occurring in a location and needs to be explained in this way does not seem to interfere with the cogency of the explanation of either the inclination's motivational or the experience's representational content.

While the content of sensory inclinations is restricted to internal states of the agent, the content of level 1 representations will include the conditions of the external world that guarantee (or make more likely¹⁴) the fulfillment of her sensory inclinations by the behaviours motivated by these representations and inclinations. An agent's level 1 representations might represent that there is an object in front of her; it looks cherry-like (it is small, round, red, etc.); if it looks cherry-like then it will taste cherry-like; the cherry-looking object is not hard; if the object is hard then chewing it will be painful; etc. The agent could be wrong about any of this: she could be hallucinating the object, expecting an olive that feels cherry-like to look cherry-like, expecting a cherry-looking cherry tomato to taste cherry-like, expecting a cherry-looking painted rock not to be hard, etc. If so, then her representations and inclinations will combine to cause behaviour that would fulfill the inclinations if these conditions obtained, but will, for some inclinations and eventualities, fail to fulfill them.

Since the contents of an agent's level 1 representations are the conditions under which these behaviours yield outcomes that reinforcingly satisfy her sensory inclinations, her level 1 representations are made correct or incorrect by the external world in a familiar sense in standard environments. But level 1 representations make few commitments about the underlying nature of the world, as is plausible for basic perceptions or perceptual beliefs. Suppose that the agent's brain were removed from her body, placed in a nutrient vat, and connected to a stimulation program that involves the exact same contingencies between her behaviour (understood as motor command output) and those sensory states of her brain that reinforcingly satisfy her sensory inclinations. Her level 1 representations would now be made correct or incorrect by features of the program instead of facts about the interaction of her body with other medium-sized objects.¹⁵

An agent must ascend farther up the hierarchy of iterations of F and R in order to represent or care about the difference between entities that are fully experientially indistinguishable, like whether she is eating genuine or imitation cherries, or whether the local watery stuff is H₂0 or XYZ. To do this, she must along the way develop something like representations or motivations that concern whether certain things are caused by other things, and whether various of her experiences have similar causes (see, e.g., Chalmers, 2012, pp. 312–378). We leave the investigation of what success semantics should say about how exactly this works to another occasion. But we hope that we have said enough here to make clear how, by using reinforcing satisfaction to explain the content of an agent's inclinations for her own sensations, success semantics can, as Whyte hoped, successfully lay a foundation for explaining the content of her other mental states.